For a number of titles (I don't know how many), the metadata wrangler has bad thumbnail URLs. The 'correct' thumbnail URLs lead to zero-byte image/jpeg files.

A good example is this title:

http://metadata.librarysimplified.org/lookup?urn=urn:isbn:9781101099650

The full-size cover image is hosted at the new site, covers.nypl.org.

http://covers.nypl.org/Content+Cafe/ISBN/9781101099650/Jacket.jpg

The thumbnail is incorrectly given as being at the old site, book-covers.nypl.org:

https://s3.amazonaws.com/book-covers.nypl.org/scaled/300/Content+Cafe/ISBN/9781101099650/cover.jpg

It exists as a zero-byte file on covers.nypl.org:

https://s3.amazonaws.com/covers.nypl.org/scaled/300/Content+Cafe/ISBN/9781101099650/cover.jpg

So there are a variety of problems on display here. book-covers.nypl.org is somehow sticking around as the thumbnail site, the thumbnail isn't actually being uploaded to the correct place, and the thumbnail site is staying in the verbose s3.amazonaws.com form despite the fact that the metadata wrangler has a bucket_name_transform configured.